Enhanced method and apparatus for deducing a correct rail weight for use in rail wear analysis of worn railroad rails

Abstract

A system for identifying rail weights applicable to worn railroad rails measured along a railroad track system includes at least one track measurement device for recording at least a portion of a profile shape of the worn rails, and, a data processor running a software implement for accepting recorded data as input data from the at least one track measurement device and for processing the input data against known unworn railroad rail data to facilitate selection and application of the appropriate rail weights equated thereto to the worn railroad rails.

Description

FIELD OF THE INVENTION

The present invention is in the field of railroad maintenance and pertains particularly to methods and apparatus for deducing correct rail weights for use in rail wear analysis of worn railroad rails.

BACKGROUND OF THE INVENTION

Maintenance of railroad tracks is extremely important for safe operation of trains on railroads. Widening of gage (the separation between the left and right rails), excessive wear of rails and fatigue-induced cracks in the rails can cause hamiful consequences such as derailments. In order to avoid such incidents, it is necessary for railroad owners and authorities to inspect the track conditions at regular intervals. To acquire information about the condition of railway tracks, important track parameters are often measured by means of specialized track geometry measurement vehicles or ‘Geocars’. The track parameters thus measured include track geometry (gage, alignment, curvature, cross-level, surface quality, etc.) as well as the wear on rails.

Rail wear, in particular, is an important consideration when it comes to safety of railroad operation as well as the quality of ride. Apart from imparting a bumpy ride, an excessively worn rail may create undue stresses in the rail due to reduction in cross section, widen the gage of the track and may even cause a derailment. Rail wear is especially pronounced on curved portions of track, where the inner face of the outer (‘low’) rail has to bear the brunt of the forces exerted by the vehicle. On straight tracks (often called ‘tangents’), the head of the rail experiences wear though the rate of wear is less than that seen on curves. Therefore, it is important to know the amount of wear on rails so that maintenance activities to replace the most critically worn rails can be taken up to avoid any adverse consequences.

A typical rail cross-section is divided into three parts: The lowest part is the wider ‘base’ that serves as a support to the rail and is used to attach the rail to the ballast. The topmost part is the thicker ‘head’ that supports the wheel and is the part subjected to wear. The middle part is the slender ‘web’ that joins the head to the base.

The wear on rails mainly affects the top of the head (termed as ‘head-wear’) and the inner side of the rail (termed as ‘gage face wear’). Measurement of the shape of a rail is of prime importance in estimating the amount of wear that a rail undergoes due to the passage of trains. Track geometry measurement vehicles (often called ‘geocars’) are equipped with laser-based instrumentation to measure the transverse cross-sectional view of the rail (often called ‘rail profile’) at approximately regular intervals on the rail track. The instrumentation obtains an image of the rail, which is illuminated by laser sheets. This image is typically digitized and converted to a set of points and the corresponding X-Y co-ordinates for further use. This recorded profile of the worn rail is compared to that of a standard unworn rail in order to estimate the wear that has taken place on the rail. The above-described functionality is often available in, or as part of the on-board software in commercial wear measurement systems and an estimate of wear can be obtained by comparing the standard cross-section, belonging to the rail weight of the rail under consideration, to the recorded one.

One challenge to comparison of worn rails to their new un-worn counterpart is that there are many types of rail cross-sections typically used by railroads. These are often designated by ‘rail weights’, which in turn, relate to other mechanical properties of the rail. Smaller and weaker cross sections are used on low-tonnage tracks, while larger and stronger cross-section designs are needed for high-tonnage tracks. In addition, rail manufacturers occasionally, to take care of railroad requirements, introduce new designs of rails. As the tonnage-carrying requirement of a track evolves, the type of rail used to build the track and its cross-section also evolves. In addition, rails belonging to differing rail weights according to availability are often used to replace a particular portion of track that needs replacement. Also, occasionally, railroads undergo business consolidation mergers and acquisitions, wherein the track belonging to a railroad can be taken over by another railroad, thus resulting in a possible introduction of new rail weights to the track belonging to the acquiring railroad. Therefore, a railroad may possess a track made up of several such rail weights, where each one has a different cross-sectional profile. Typically, there can be as many as 50 to 100 different rail weights being used over the track belonging to a large railroad.

Typically, during a track geometry measurement test run, a rail weight is entered by the operator at the start of a test or sometimes during a test. However, as the test progresses, the track rail weights change and it is not possible for the operator to keep track of the changing rail weights to make accurate entries during the measurement process. The on-board software can find out wear using the standard cross-section available to it. But more often than not, it ends up comparing the recorded rail profile to the standard profile belonging to an incorrect rail weight, since the rail weight entered by the operator was erroneous. This results in frequent miscalculation of wear on rails.

Manufacturers of commercial wear measurement systems realize this and provide an easy-to-use graphical interface to visualize the recorded profiles vis-à-vis standard profiles for different rail weights. This allows the user to select profiles for different standard rail weights, compare them visually against the recorded profile and choose the appropriate standard rail weight that is the closest match to the recorded profile. This manual process is often used to correct the wrongly entered rail weights and to obtain the correct amount of rail wear on each rail. However, this is an offline process and often proves to be cumbersome and labor-intensive. Being a manual process, it is prone to human errors and takes long time to complete. More often than not, due to unavailability of manual labor and the meticulousness required for the process, the wear estimates are available several weeks after the data is recorded. It is critical to know the amount of wear on rails as soon as possible, especially in case of badly worn rails, and a delay in obtaining these estimates may even result in serious consequences such as derailments.

Commercial wear measurement systems have attempted to provide such functionality by utilizing various methods such as using similarity of geometric parameters, using business rules for identifying rail weights, identifying engraved numbers on the side of the rail, and monitoring the variation of rail shapes over a given length of rail track. The success of these techniques has been varied and it is observed that these techniques often fail to discern profiles that are very close to each other in shape, profiles that are highly worn, profiles that are completely new to the system and profiles that were recorded in presence of environmental irregularities such as adverse weather conditions, dirt, grease, snow and mud.

Also, in cases of business consolidation mergers and acquisitions, wherein the track belonging to a railroad is taken over by another railroad, entirely new rail weights, not included in the standard rail weight set known to the operator or the wear measurement system, may be introduced in the track. In such cases, the new rail weights need to be identified as ‘unknown’ or ‘foreign’ rail weights for wear calculation as well as for keeping records.

DESCRIPTION OF KNOWN PRIOR ART

Most of the prior art in the area of measuring rail wear deals with the apparatus and methods for acquiring rail profile shapes of worn rails. Some of these are also related to the area of grinding rails to required profiles. Optical as well as mechanical means have been used for the same.

The inventor is aware of numerous patents related to rail wear measurement appearing mostly in the class definitions 356, 73 and 33 of US Patent Classification index. The prior-art references found are identified and abstracted below for convenience.

A U.S. Pat. No. 4,069,590, hereinafter termed Effinger discloses an automated rail wear measurement system. It includes an assembly movable along a railroad track with displacement pickups bearing upon top surface and inner side surface of each rail. Entered into the system are unworn top and side rail surface measurements taken relative to a known unworn feature of the rail at the selected points. The pickups generate signals in accordance with the mechanical displacement of the top surface and the side surface points relative to the measured unworn top and side surface points, and thus enable calculation of wear.

A U.S. Pat. No. 4,417,466, hereinafter termed Panetti discloses both method and apparatus for measuring at least one geometrical characteristic of the transverse profile of the head of at least one rail of a railway track. It uses a reference base perpendicular to the axis of the track and parallel to a tangent line to the rolling surfaces of the two lines of rails. It determines the distances separating at least two longitudinal sidelines of the tracing of the head of a rail from the reference base. The reference base may be represented by part of a measuring carriage rolling on the track by means of rollers resting vertically and horizontally against the two lines of rails.

A U.S. Pat. No. 4,577,494, hereinafter referred to as Jaeggi discloses an apparatus and method for measuring rail wear. It consists of a chassis adapted for travel along a railroad rail with multiple bearing points located between the chassis and the rail for supporting the chassis on the rail and for allowing movement of the chassis over the rail. The relative vertical displacement of the bearing points allows calculation of wear.

A U.S. Pat. No. 4,625,412, hereinafter referred to as Bradshaw discloses an apparatus and method for measuring rail wear. It consists of optical sensors mounted to a section of a carriage adapted for movement along a rail. Wear in the profile of the rail causes movement of the carriage, and the optical sensors detect a change in the position of the carriage relative to at least one of the unworn surfaces of the rail. By dividing movement of the carriage section into vertical and horizontal movement, the distance sensors provide an indication of wear.

A U.S. Pat. No. 4,915,504, hereinafter referred to as Thurston discloses an optical wear measurement system. It uses CCD cameras to capture image of the rail illuminated by lamps. The image is processed to locate key geometric points on the profile such as center point of the web and the base, which are used to find the wear on the rail.

A U.S. Pat. No. 5,140,776, hereinafter referred to as Isdahl, et al discloses an apparatus and method for measuring and maintaining the profile of a railroad track rail. It uses a strobe light, oriented directly over a railroad track rail at a 45° angle to the horizontal, which projects a light line across the rail. A pair of cameras oriented directly over the rail view the light image on the rail. The light lines reflected from the foot of the rail provide a reference for numerically describing the profile of the railhead. The measured rail profile is compared to an ideal rail profile. The results of the comparison are used to position grinding modules for grinding the rail to a preferred profile.

A German patent DE 10040139, hereinafter referred to as Worbs, et al discloses a device and method for measuring rail wear using inertia measurement systems in conjunction with imaging systems. The imaging system measures the relative distance between the real position of the rail course in space to the reference position obtained using the inertia measurement system, and it also measures the contour profiles for the running edges of the left and right rails.

A European patent EP00007227, hereinafter referred to as Steel, et al discloses a method and apparatus for measuring the profile of an elongated surface. Surface profile measurement is carried out, where there is relative movement between the observer and the surface, by beaming a fan-like, planar beam of light on to the surface at an acute angle, and monitoring the pattern of the reflected light. In its application to measurement of rail wear on a railway track, the pattern of the light reflected is compared with the pattern of light reflected in a similar geometrical arrangement, from an unworn rail.

The inventor has studied the above-mentioned documents and has found that they do not solve the problems of the prior art. In general wear calculation as is known in the art relies upon one of or a combination of the following:

• • The requirement for finding displacement of measuring mechanical members
  • The requirement of capturing optical images of worn rail shapes
  • The requirement of measuring geometric distances between certain key points
  • The requirement of measuring a parameter (such as shape, distance, pattern of reflected light) of a worn rail and comparing it against a known standard, but where the standard is invariably specified by an external agency or is constant and manually selected.

Nothing in the art listed above solves or even adresses the problem of manual rail weight identification and system entry being required in order to successfully and accurately obtain all of the parameters for accurate rail wear measurement.

In addition to the extensive list of references provided above, the inventor is further aware of a U.S. Pat. No. 6,356,299, hereinafter referred to herein as Trosino, et al. Trosino, et al discloses an automated track inspection vehicle and method for creating images of the track and detecting anomalies. The inventor has found that the only portion of Trosino that might be remotely considered relevant to the art of this specification is contained in the background description of Trosino and is herein reproduced verbatim below:

“In addition to these automated inspection systems, pattern recognition systems are beginning to be utilized in railroad applications. One example is a rail profile measuring system, in which a video camera is utilized to view the rail and measure the shape of the rail. The images are returned to a computer to identify defects in or excessive wear of the rail. Additionally, the system employs a pattern recognition algorithm to compare the image of the rail to a preselected database of rail shape to identify the particular type of rail measured.”

While the reference does mention in an un-enabhng fashion the possible existence of systems using pattern recognition methods to identify an original shape of a rail, it provides no solution to the problem of manual entry of rail weights and associated un-worn profiles of those rails for use in comparison.

In addition to the above material references the inventor has also become aware of a thesis by Todd Wittman (University of Minnesota), which discloses a method based on geometric moments to automatically identify the best match for a measured rail profile from a set of reference profiles for rail grinding. However, the set of reference profiles used here consists of worn profiles and not of unworn standard profiles. This is not the same as the problem described in the background section, wherein the best match needs to be identified from a set of standard unworn profiles, given a recorded profile that could be highly worn or relatively new. The reference to the thesis is provided herein as

• http://www.math.umn.edu/˜wittman/dinah/Msthesis.wrd.

The inventor is aware that certain commercial systems possess the ability of automatic rail recognition. The manufactures of such commercial systems include the following:

Industrial Metrics, Inc., USA

• http://www.industrialmetrics.com/
  Formerly, Range Vision, this Canadian company manufactures a suite of products for measurement of rail wear (‘RangeCam’ products). Incorporates use of automatic rail weight recognition.
  KLD Labs, USA
• http://www.kldklabs.com/
  Manufactures a suite of products for measuring track geometry including rail wear. Incorporates use of automatic rail weight recognition. Also makes use of the solution by Industrial Metrics referenced immediately above.
  Holland Co., USA
• http://www.hollandco.com/
  Manufactures a suite of products for measuring track geometry including rail wear. Incorporates use of automatic rail weight recognition. Also makes use of the solution by Industrial Metrics.
  ENSCO, USA
• http://www.ensco.com/
  Manufactures a suite of products for measuring track geometry including rail wear. Incorporates use of automatic rail weight recognition. Method used for the same not known. No patent information available.
  ImageMap, USA
• http://www.imagemap.com/
  Manufactures a suite of products for measuring track geometry including rail wear. Also makes use of the solution by Industrial Metrics.
  Electrologic, Australia
• http://www.electrologic.com.au/
  Manufactures a suite of products for measuring track geometry including rail wear. Incorporates use of pattern recognition and image analysis technologies, but not necessarily used for automatic rail weight recognition.

According to the inventors, it is reported that the aforementioned commercial systems experience difficulty in discerning profiles, which either are very close to each other in shape, or which are highly worn, or which are completely new to the system, or which were recorded in adverse weather conditions. Although these commercial systems claim the ability to deduce rail weights in addition to standard un-worn rail profiles from measuring worn rails in the field, the inventors believe these systems use a set of business rules and/or certain geometric parameters for making the identification.

A drawback to this approach is that if business rules and/or measurement of key geometric parameters without the use of sophisticated statistical techniques aimed at generalization were used in a method employed for identification of rail weights, it is very likely that many such rules would be required to discern a huge number of rail weights from each other. Any alterations to the standard set of rail weights such as addition of a new rail weight are likely to demand a significant change in the set of governing business rules. In addition, though the variation encountered in a particular geometric parameter being considered for identification, can be minimal for newly laid rails, it can vary considerably over the entire track for older rails. It is a tedious task to accommodate such variation in a set of rules.

Therefore, what is needed in the art to solve the above-stated problems is an automated system and method that enables a user to deduce the correct rail shape and weight of an un-worn rail that corresponds to and is deduced from a worn profile of a rail measured. Such a system could discern subtle differences from closely-resembling rail weights; work in cases of highly-worn rails as well as relatively new rails; identify a rail weight that is unknown to a provided set of standard rail weights used for comparison; and enable successful identification and wear calculation even in the event that part of a rail profile measured is not properly recorded.

SUMMARY OF THE INVENTION

A system is provided for identifying rail weights applicable to worn railroad rails measured along a railroad track system. The system includes at least one track measurement device for recording at least a portion of a profile shape of the worn rails, and a data processor running a software implement for accepting recorded data as input data from the at least one track measurement device and for processing the input data against known unworn railroad rail data to facilitate selection and application of the appropriate rail weights equated thereto to the worn railroad rails.

In a preferred embodiment, the track measurement device uses image-capturing technology to record track data. In another embodiment, the track measurement device further includes audio emitting and capturing technology to record track data. The appropriate rail weights are identified for use in rail wear measurement of the physical wear subjected to the worn rails.

In one embodiment, the at least one-track measurement device is installed in a track measurement railcar. In a preferred embodiment, the profile shape of the worn rails is a transverse cross-sectional view of the rail shape. Also in a preferred embodiment, the software implement directs rail weight identification and rail wear calculation. In all embodiments, the profile shape of a rail comprises a set of geometric features applicable and identifiable to parts of the rail profile.

In a preferred embodiment, the geometric features include one of, all of, or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle. In this embodiment, the unworn rail data comprises profile shapes of unworn rails, each profile including a set of geometric features making up the profile shape, the features applicable and identifiable to parts of the rail profiles. Likewise in this embodiment, the geometric features include one of all of or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle.

In a preferred embodiment, the system further includes a data repository for storing calculated rail-wear data results along with geographic information and date and time information.

According to another aspect of the present invention, a software implement is provided for identifying rail weights applicable to worn railroad rails measured along a railroad track system. The software implement includes a feature extractor for extracting geometric features from recorded data of the worn rails, and a rail weight classifier for comparing the extracted features to the same features of unworn rail shapes and for selecting an appropriate rail weight equated to an unworn rail shape for application to a worn rail.

In a preferred embodiment, the feature extractor combines geometric feature data produce a geometric vector. Also in a preferred embodiment, the rail weight classifier compares the geometric vector against a store of geometric vectors previously created for unworn railroad rails. In this embodiment, the geometric features derive from captured image data. In an alternate embodiment, the geometric features derive from captured audio data.

A preferred embodiment, the software implement further includes a preprocessor for creating X, Y coordinate pairs for feature extraction. In still another embodiment, the software implement further includes a post processor for validating identified rail weights applied to worn rails against a set of business rules. In yet another embodiment, the software implement includes a rail wear calculator for calculating the correct amount of wear subject to a worn rail compared to an unworn rail of the same identified rail weight.

According to another aspect of the present invention, a method is provided for identifying a correct a rail weight for a worn rail profile given data recorded from the worn rail by a track measurement system. The method includes steps of (a) identifying geometric features of the worn rail profile; (b) unifying the identified features as a single feature vector; (c) comparing the single feature vector to one or more similarly constructed feature vectors previously stored for comparison, the previously stored feature vectors representing unworn rail profiles; and (d) selecting a feature match of that of a worn rail profile to that of an unworn rail profile.

In a preferred aspect in step (a), the geometric features of the worn rail profile include one of, all of, or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle. In this aspect the geometric features are derived from image data captured from a worn rail. In another aspect, the geometric features are expressed as a set of X, Y coordinate pairs.

In one aspect in step (a), the geometric features are discernible from one another using linear or quadratic discriminant analysis. In a preferred aspect in step (b), the single feature vector contains the data from all of the unified features considered in creating the vector.

In one aspect in step (d), if there is no match accomplished to a previously stored feature vector, an indication of match to a rail weight not known to the system is returned. In another aspect of the method in step (a), the identified geometric features are adjusted to correct for a detected angle of cant of the worn rail. In still another aspect, the method further includes a step (e) upon obtaining a match in step (d) wherein the resulting data of the match is input into a rail wear calculator.

According to yet another aspect of the present invention, in a system for identifying rail weights applicable to worn railroad rails measured along a railroad track system, a method is provided for correcting a worn rail profile for an angle of cant. In a preferred embodiment, the method includes steps of (a) detecting the existence of cant attributed to a web portion of the rail profile; (b) retrieving a sample of any like web portion of an unworn rail aligned to true vertical; (c) cross-correlating the web portion of the worn rail to the web portion of the unworn rail; (d) identifying the direction of cant of the worn rail against the unworn rail; (e) rotating the web portion of the worn rail a number of increments of angle opposite the cant direction; and (f) accepting as a correction angle, the sum of rotation increments required to obtain a least discrepant match in cross-correlation to the symmetry of the web portion of the unworn rail profile.

A preferred aspect in step (a), the worn rail profile is described by a set of X, Y coordinate pairs. In one aspect in step (a), more than one worn rail profile is considered for cant correction along a same track section.

In a preferred aspect in step (b), the sample is retrieved from a repository previously stored unworn rail profiles. Also in a preferred aspect in step (c), cross-correlation considers vertical symmetry. In another aspect in step (c), cross-correlation considers horizontal symmetry. In a preferred aspect in step (c), cross-correlation is two-dimensional.

In one aspect, the method further includes a step (g) for estimating height of the worn rail profile. In one aspect in step (f), a least discrepant match considers the best equation for a line of symmetry of the worn rail profile considering the minimum distance equalities of like geometric features located on either side of the line of symmetry in cross-correlation.

In a preferred aspect in step (d), selection is based on pattern recognition results. In this aspect pattern recognition is based on linear or quadratic discriminant techniques and associated rules. In another aspect, pattern recognition is based on use of a decision tree.

In one embodiment, with reference to the system described further above, the processing includes statistical comparison. Likewise with respect to the software implement, the method for comparing is statistical comparison. Also in one embodiment with reference to step (c) of the method for identifying rail weights, the method for comparing is statistical comparison.

BRIEF DESCRIPTION OF THE DRAWING FIGURES

FIG. 1 is a plan view of a typical rail profile 100 as is known in the art.

FIG. 2 is a plan view of a rail profile belonging to a worn rail superimposed over a rail profile belonging to a standard unworn rail type of different rail weight.

FIG. 3 is a plan view of a rail profile belonging to a worn rail superimposed over a rail profile belonging to a standard unworn rail type exhibiting subtle differences in like geometric parameters.

FIG. 4 is a graph illustrating mathematical separation between plotted X, Y coordinate sets of a worn rail pertinent to certain geometric features compared to those of an unworn rail pertinent to the same features.

FIG. 5 is a block diagram illustrating a component for rail profile comparison and rail weight classification according to an embodiment of the present invention.

FIG. 6 is a plan view of a rail profile illustrating certain features that may be used in comparison to find a correct rail weight according to an embodiment of the present invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

In accordance with embodiments of the present invention, the inventors provide methods and apparatus for automatic identification and selection of a correct rail weight for a worn rail using the worn rail profile as data input. The methods and apparatus of the present invention will be described in enabling detail below.

As was described in the background section above, prior-art methods for rail-wear calculation are adept at acquiring the shape of worn rails, which is the primary reason for their use in rail wear measurement. However, they may rely on user-specified rail weights for calculation of wear. Since in actual practice, the track is made up of several differing rail weights, a separate manual process involving visual comparison of rail profile shapes with that of standard profiles may be required to establish correct rail weights and calculate wear accurately. Some of the commercial systems listed in the background section above claim to possess automatic rail weight recognition capabilities of identification of the original rail weight given the shape of the worn profile, but are believed to suffer from some of the following deficiencies:

• • Inability to discern rail weights which are very close to each other in standard profile.
  • Inaccuracy in identification of profiles, which are highly worn.
  • Inability to detect profiles of rail weights that are completely new (‘foreign’ rail weights) to the system.
  • Inaccuracy in identification in case of profiles recorded in adverse weather conditions.
  • Lack of consideration for statistical variation in geometric parameters.

Therefore, a goal of the present invention is provision is to provide an automated system that is capable of:

• • Identification of the unworn standard rail shape and rail weight of a worn rail shape under evaluation by comparing the geometric shape of the worn rail to that of standard unworn rails
  • Presenting the shape of a worn rail as a set of curves, lines, areas, angles and other like geometric features and comparing these features with same features of standard rail shapes
  • Utilizing key differentiating geometric features in comparison, which make a particular rail shape different than other rail shapes
  • Utilizing certain differentiating geometric features in comparison, which do not change considerably with time due to the effect of rail wear.
  • Utilizing certain geometric features in comparison, which may not be easily discernible by visual inspection.
  • Utilizing many such geometric features, so that if some features are not recorded properly due to conditions such as weather or instrument errors, the remaining features can be used for comparison.
  • Comparing all relevant geometric features of a worn rail shape to like features of standard rail shapes in comparison and deciding which standard rail shape, if available, provides the best overall match and if unavailable, provided indication of a foreign rail shape not known to the system.
  • Providing a quantity of measure for the degree of match of the worn rail shape with the identified standard rail shape in comparison
  • Incorporating practically encountered statistical variation in geometric parameters being considered for identification.
  • Creating and updating a rail-lay database describing rail weights as a function of geographic location on the track.

FIG. 1 is a plan view of a standard unworn rail profile 100 according to typical art. Rail profile 100 represents a cross-sectional profile of a typical railroad rail used in a typical railroad track segment. Rail profile 100 has three basic geometric components. These are, a railhead 101, a rail web 102, and a rail base 103. Rail base 103 is the widest portion of profile 100 and is the portion used to attach the rail to a ballast as is known in the art. Railhead 101 is the portion of rail profile 100 used to support the wheels of a train car and is the portion that is subject to the most physical wear. Rail web 102 is the slender support portion of profile 100 used to connect head 101 to base 103. Although in actual practice, a profile taken from a laid rail may exhibit a certain lean with reference to a vertical center, rail profile 100 may be assumed in this rendition to be vertically aligned or vertically adjusted to be represented symmetrically on a true vertical center-line.

It can be seen in this example that certain geometric features are exhibited by the components of rail profile 100. For example, railhead 101 is slightly rounded at its corners, and may not have absolutely parallel sides. That is to say that the left and right sides of rail head 101 may, in relation to each other, form a negative or positive taper defined at each side as an angle measurement from vertical. Railhead 101 may be thought of as a broad feature of rail profile 100, the feature that would most likely receive the most physical wear during actual use.

Physical wear on a rail, as previously described, largely affects a railhead such as railhead 101. There are two types of wear typically seen on railhead 101. For example, wear affecting the top of railhead 101 is termed head-wear. Physical wear affecting, typically, the inner side of railhead 101 is termed gage face wear.

As alluded to above, many commercial wear measurement systems attempt to capture the image of the shape of a worn rail at a transverse cross-section of the rail by employing optical means. The shape of a worn rail thus recorded is often called a ‘rail profile’, hence termed herein rail profile for element 101. Though a worn rail shape can represent the complete shape of the rail it is often available only as a partial description of the rail due to equipment limitations of measurement systems known in the art. Typically, the measured rail shape represented by known systems consists only of the head part and the web part. The part of the rail joining the head and the web as well as the entire base part are not typically recorded due to peculiar placement geographies of the recording elements of known systems. The present embodiment assumes such a worn rail shape in rail profile 100, however other profile descriptions can be used in place of a transverse cross-section sectional profile without departing from the spirit and scope of the present invention.

In a preferred embodiment of the present invention, the shape or profile of a worn rail analogous to rail profile 100 is described, at least as strategic locations, by a set of pairs of X-Y coordinates (not illustrated). X-Y coordinate pairs may be obtained by various methods including digitization of an image of the worn rail produced by a commercial wear measurement system. Commercial wear measurement systems provide a means of producing such rail ‘profiles’ mainly for the purpose of visualization, which are often available with majority of railroads employing such systems. Though in the preferred embodiment of this invention, such X-Y profiles are assumed to be readily available, they can be generated given the images of worn rails using standard image processing algorithms in various embodiments. The preferred embodiment considers that a set of pairs of X,Y coordinates describing the shape of the worn rail serves as input data used to produce corresponding rail weight identification, which may then be used to calculate rail wear. In another embodiment geometric features may, in some cases, be derived from the use of sophisticated sound emitting and echo receiving equipment, although in the preferred embodiment geometric features are produced from image data. The present invention incorporates a methodology, described in more detail below, and based on statistical measurement of how closely a worn rail shape matches with an unworn standard rail shape. A worn rail shape, analogous to rail profile 100 is presented as a composition of various geometric parameters or features including, but not limited to, curves, lines, radii, areas, angles and distances. Careful examination of rail shapes belonging to a variety of standard rail weights reveals that certain geometric parameters can be used as distinguishing features, which can authoritatively differentiate a particular rail shape from other rail shapes.

FIG. 2 is a plan view of a rail profile comparison 200 of a rail profile belonging to a worn rail superimposed over a rail profile belonging to a standard unworn rail type of a differing rail weight. Profile comparison 200 is logically illustrated in this embodiment as a rail profile of a worn rail (dotted line profile) superimposed over a standard rail profile (solid line profile) of an unworn rail known to the system.

For example, consider the separate profiles also termed rail weights 201 (wom profile) and 203 (unworn profile) as shown in this example. In comparison of profiles 201 and 203, there are two distinct profile height dimensions exhibited, H (1) and H (2). Understandably, the heights are different owing to a fact that one of the profiles is of a worn rail, which may have head-wear that significantly reduces the rails height.

For the purpose of discussion the overall height of a rail profile is measured from the bottom of the base portion (not illustrated herein) to the top of a railhead at the highest point. It should be noted herein, that some rail profiles of unworn rails may have a mere flat surface area on top of the railhead, wear as some rail profiles may exhibit a slight radius to the surface wherein the height dimension decreases naturally toward the left and right edges of center in a transverse cross-sectional profile. In one embodiment of the present invention there may be more height measurements involved in comparison of the two dimensions of this example, H (1), and H (2).

Considering profile height dimensions alone in a method to find the correct unworn profile to match a worn profile could produce unreliable results if a rail is significantly worn such that its height has been significantly reduced. Therefore, it may be difficult to determine whether a worn profile 201 belongs to the same rail weight as profile 203. Visually speaking, the rest of the profile comparison from the top part of the rail web down to the base appears identical. However, the standard geometry of the super imposition indicates that width of the head of profile 201 is discernibly greater than that of the head of profile 203. In this example, head width is a distinguishing feature at lends to the correct assumption that the rail weights involved in the comparison of this example are different rail weights. While the rail weight of profile 201 is still unknown it is assured that it is not the rail weight of the standard rail profile 203 being compared thereto in process.

FIG. 3 is a plan view of a rail comparison 300 of a rail profile belonging to a worn rail superimposed over a rail profile belonging to a standard unworn rail type of a different weights, illustrating differences in like geometric features. In this example, a standard unworn rail profile 301 is being compared against a worn rail profile 302. In this particular example the compared profiles include the features rail height, railhead widths W1 for rail profile 302 and W2 for rail profile 301. In addition to the just-mentioned features, to fillet radii 304 for rail profile 302 and fillet radii 303 for rail profile 301 are compared.

In this example the heights are relatively close, while W2 is discernibly smaller than W1 and fillet radius 304 is discernible a different and fillet radius 303. As was described above with reference to FIG. 2, head width is a distinguishing feature. Likewise, fillet radii joining the web portions of the profiles to the base portions of the profiles (303 and 304) are distinguishing features. In a preferred embodiment of the present invention, at least a single or a combination of multiple distinguishing features may be leveraged in a comparison process of a rail analyzed in the field against a database of standard unworn rail types in order to determine the correct and rail type and weight of the rail analyzed in the field. By utilizing this technique, automated rail wear calculation may proceed without intervention by human operator in selecting and inputting any particular rail weight or weights for comparison. Such distinguishing features generally experienced little or no wear in the field and therefore provide reliable data for comparison.

It will be apparent one with skill in the art, that more than one geometric feature of any rail analyzed in the field may be compared against like features of one or more rails known to the system without departing from the spirit and scope of the present invention. Although the comparison examples of FIGS. 2 and 3 illustrated a single worn rail profile compared to a single standard rail profile, it will be clear to one with skill of the art that one rail analyzed in the field may be compared to all of the rail profiles known to the system in a single pass. Using X,Y coordinates pairs, various like geometric features or like combinations of geometric features of candidate profiles for comparison may be plotted on a graph, which may be used to discern even subtle differences between the features that would not otherwise be discernible to a system.

FIG. 4 is a graph 400 illustrating mathematical separation between plotted X,Y coordinate sets of a worn rail pertinent to certain geometric features compared to those of an unworn rail pertinent to the same features according to an embodiment of the present invention.

As was described further above with reference to FIG. 3, a single geometric feature or a combination of such features measured and then plotted as a set of X,Y coordinate pairs, in this example on graph 400, helps to distinguish a rail shape from other rail shapes. In this example, the head widths W1 and W2 of profiles 302 and 301 respectively are plotted against their fillet radii features 304 and 303 respectively. Graph 400 plots fillet radii from 0.5 inches to 1 inch. Graph 400 also plots head width from 2.8 inches 3.1 inches. It can be seen that isolated ‘clusters’ are formed of the X,Y pairs, clearly distinguishing the two rail profiles other comparison from each other. A line dividing these two separate clusters, which is termed a linear discriminant, can be located. The linear discriminant may be used to make a decision on which rail weight from a database of rail weights a worn rail shape belongs to given the head width and height calculated from the worn rail shape.

Therefore in a preferred embodiment, the technique of statistical discrimination is leveraged in comparison and the line dividing the clusters of X,Y pairs is the linear discriminant. If many different rail weights are under consideration, many such clusters are formed and a curve instead of a line is used to separate those clusters. Such a curve is termed a quadratic discriminant, and has been used in accordance with the present invention for identifying original rail weight of a worn rail shape. The present example considers only two features and therefore can be visualized as a 2-dimensional plotting. However in a preferred embodiment many such geometric features are considered in the process of the present invention for effective identification of an overall match between a worn rail and a standard unworn rail.

FIG. 5 is a block diagram illustrating a software component 500 for rail profile comparison and rail weight classification according to an embodiment of the present invention.

FIG. 6 is a plan view of a rail profile 600 illustrating certain features that may be used in comparison to find a correct rail weight according to an embodiment of the present invention.

Referring now to FIG. 5, software component 500 may accept at least one rail profile 501 as data input. Data about worn rail profile 501 comprises X,Y coordinate pairs describing at least one, but more likely, several features of a worn rail profile being analyzed for comparison. Such features may include, but are not limited to head width, overall height, web width, web radius, fillet radius, and flange angle. More detail about such geometric features will be provided later in the specification.

Data collected in the field from a worn rail profile aggregated as input data 501 is input into a preprocessor 502. Preprocessor 502 accepts the input data and prepares it for further processing. A rail shape recorded with automatic optical instrumentation may often be subject to various optical recording inconsistencies. For example, an optically recorded shape may show additional points plotted away from the cluster of points of the actual rail shape as seen in a graph such as graph 500 of FIG. 5. Reasons for this may vary but may include badly recorded edges, effects caused by interfering rail parts, or perhaps poorly recorded shapes caused by weather problems. Moreover, as previously described, rail shapes may only be partially recorded wherein the portion joining the head to the web and perhaps a portion of the base is missing.

Still further to the above, a worn rail in the field may be slanted at some angle away from vertical either by design or due to the load of passing trains in some track portions, often termed a cant in the art. The cant needs to be corrected before the worn rail shape can be compared to standard rail shapes. Preprocessor 502 performs all of the required corrections to the data before any comparison process ensues.

In one an embodiment, preprocessor 502 eliminates stray points by applying a pre-known minimum distance between successive points of a worn rail shape. In a preferred embodiment processor 502 further determines whether an adequate number of points have been recorded.

Preprocessor 502 performs calculations to separate the coordinates of the worn rail shape from those of the rail had and those of the web portion of the rail. In one embodiment, ordinance from the base portion of the rail may also be included. Once the coordinates are separated preprocessor 502 performs a re-sampling of the entire shape at uniform intervals for facilitating calculation of geometric features.

In estimating and correcting for a cant in a rail profile, the system of the present invention may use a variety of methods. In one particular method, the web portion of a worn rail is divided into left and right portions on either side of a vertical centerline. Therefore, for each point recorded on the left side of the web portion, a corresponding point is located and recorded on the right side of the web portion. A minimum distance factor or rule may be applied here as described further above. In this method the recorded points for pairs, which in turn provide a set of lines that are parallel to the base of the worn rail shape.

An equation of a line is then passed through the mid points of these pairs, the equation determined by use of a suitable curve fitting technique. In a preferred embodiment when using this particular method the least squares curve fitting algorithm used for the purpose of determining the mid-point line. The determined line is considered then to be the recorded centerline of the worn rail shape. The slope of this line is compared with a true vertical line to find the exact angle of cant of the worn rail shape.

In another embodiment, a similar technique is employed wherein the center points of a left and right fillet radius are located using a suitable circle-fitting algorithm and recorded. A line is then fitted between the two points intersecting the points. Inclination of the line is then measured over its distance and compared with a true horizontal. In a preferred embodiment using this technique, a preferred circle-fitting algorithm used is given by Gander et al, 1994. The reference is included herein as (W. Gander, G. H. Golub, R. Strebel, Fitting of Circles and Ellipses, BIT 34 (1994), pp. 558-578. Gander, W., Golub, G. H., & Strebel, R. (1994). Fitting of circles and ellipses: least squares solution. TechReport Departement Informatik, ETH Zurich. ftp.inf.ethz.ch/doc/tech-reports/2xx/217.ps).

Still another method that may be used is iterative two-dimensional cross-correlation between the web portion of the worn rail shape and that of any unworn standard rail shape, where the angle of tilt of the worn rail shape is iterated over a limited range. The angle at which the best cross-correlation is obtained is selected as the angle of tilt or cant. Also, the line of symmetry of the unworn standard shape after obtaining the best cross-correlation also signifies the equation of centerline of the worn shape. This method assumes, of course, that preprocessor 502 has access to data of at least one unworn standard rail shape. Of the three methods of cant estimation described immediately above, one of, a combination of, or all of them may be employed in a single pass without departing from the spirit and scope of the present invention.

In a preferred embodiment of the present invention, cant is removed by subjecting the recorded coordinates to a rotation matrix R. This function performed by preprocessor 502 rotates the recorded coordinates by the same angle as the calculated cant except in the opposite direction. The following expression is given for R: $R=\left[\begin{array}{cc}\cos \left(\theta \right)& -\sin \left(\theta \right)\\ \sin \left(\theta \right)& \cos \left(\theta \right)\end{array}\right]$

Once the recorded data has been properly prepared by preprocessor 502, it may be passed to a profile feature extractor 503. Feature extractor 503 calculates the appropriate features from the data passed to it by preprocessor 502.

The geometric features described thus far in a preferred embodiment are derived from analysis of standard rail features of rail profiles common to the United States. However, like features from rails used in other countries as well as additional geometric features and/or external parameters not described herein may be incorporated in rail weight identification without departing from the spirit and scope of the present invention. The features mentioned are sufficient for notification the most common rail types. It is reminded herein that the features selected, in a preferred embodiment, include those features that do not exhibit wear with continued passage trains. As described above, such features are termed distinguishing features that are leveraged in determination of a proper will rail weight for a worn rail.

Referring now to FIG. 6, rail profile 600 exhibits just some of the features that may be used in determination of a correct rail weight of a worn rail so that accurate wear calculation can be performed. Head width is one such feature that may be described as a distinguishing feature. This is the width of the head part of the rail. By design, it can be uniform across the height of the head for certain rail weights, while it can either increase or decrease with height depending on design. Therefore, head width is often calculated at different heights. For worn rails it can also vary in field depending on the amount of gage face wear on the rail and excessive material flow on the side of rails, which is known in the art as lip.

Web width is the minimum width dimension of the web portion of a rail profile. Web width as calculated from an image can vary in the field for a same rail type due to grease, dirt build-up, ice, and interference from other track components while capturing the image.

Web radius is a dimension describing the radius of either side of the web portion of a rail profile. For the purpose of the present invention there is termed a right web radius and a left web radius. The web radius is not as subject to image error in image capturing, as is the web width, because of the reconstructive nature of a larger radius as a whole.

Fillet radius is a much smaller radius located at the bottom and of a web portion where it adjoins the base of a rail profile. As with web radius, fillet radius is a dimension describing the radius of either side and therefore there are terms for right fillet radius and left fillet radius.

Flange angle is the amount of angle that the sides of the head of a rail form with a true vertical. By design, it may be zero for certain rail weights, while it may be either a positive or negative angle measurement for other rail weights. It may vary in field depending on the amount of gage face wear (lip) that exists on a worn rail and any excessive material flow to either or both sides of a worn rail.

The height feature refers to the height of a worn rail shape from the base center to the top of the head. It may be calculated along the centerline of the worn rail shape. It may vary for a particular worn rail type in field depending on the amount of head-wear subjected to the rail.

Referring now back to FIG. 5, feature extractor 503 calculates the aforementioned geometric features and compares them against the same features of profiles stored in repository 504. To further illustrate, head width is calculated by considering the distance of the worn rail shape centerline from sides of the head part. In case of rails laid on curved tracks, the inner side of the rail is typically worn more excessively then the outer side and therefore, head width alone may not provide sufficient data for rail comparison. To improve accuracy in measurement of a profile from a curved track, the distance from centerline of the head side that is the least worn is taken from near the bottom part of the head may be calculated and then doubled to give an accurate representation of head width used for identification purposes.

Feature extractor 503 calculates web width simply by measuring the least distance or minimum distance that exists between the right and left sides of the web portion. The right and left web radii are calculated, in a preferred embodiment, using the circle-fitting algorithm described further above wherein X,Y pairs are considered. The right web radius is calculated using X,Y pairs from the right side as input, while the left web radius is calculated using X,Y pairs from the left side as input. Feature extractor 503 calculates fillet radius both right and left using the same or a similar circle-fitting algorithm.

Feature extractor 503 calculates flange angle by quantifying the slope of the head portion of a rail profile in relation to the vertical sides of the head portion of the profile. Although both left and right slope measurements are taken, calculation for flange angle for a profile considers the slope of the least worn side from centerline.

Feature extractor 503 also calculates the overall height of a rail shape, however for calculating the height, it is desired that the worn rail shape be located correctly in X-Y coordinates so that its base portion corresponds to the base portion of a standard rail shape or shapes under consideration. The inventor has observed that certain points that are available in all worn rail shapes can be fixed irrespective of the wear on the rail and can therefore be used as distinguishing points for the purpose of locating the worn rail shape correctly with respect to an unworn standard rail shape. These points include, but may not be limited to, the center points of left and right fillets as well as the midpoint of the web where the web is of the thinnest dimension.

As has been repeatedly described, the web portion of a rail may not be completely recorded by optical instrumentation and therefore the exact midpoint of the web may not always be available. The fillet portions of a web are available more often. A particular embodiment of the present invention uses feature extractor 503 to find a correct location of a worn rail shape with respect to a standard unworn rail shape by the first locating the center point coordinates of the right and left fillet and then ‘pinning’ the located coordinates on the center point coordinates of a standard unworn rail shape. This technique is called ‘pinning and refining’ in the art and involves certain steps listed below.

• • 1. The center point coordinates of the fillet part are located by using, in a preferred embodiment, the geometric circle-fitting algorithm mentioned above.
  • 2. The pinning procedure consists of determining the translation required to equate the fillet center point of the fitted circle to that of a given standard rail shape and translating the worn rail shape coordinates by that translation.
  • 3. The fitted center point is expected to be approximate and therefore, further refinement is necessary to locate the worn rail shape correctly. To accomplish this, a least-squares algorithm is employed, in a preferred embodiment, which adjusts the position of the web portion of the worn rail shape with respect to that of the standard rail shape, so that the sum of squares of the distance between the two web portions is minimum.

In another embodiment, two-dimensional cross-correlation using web portions of both the worn and the standard rail shapes sampled at the same granularity is used to superimpose the worn rail shape correctly over a standard rail shape.

In a preferred embodiment using the technique described above, the centerline of the worn shape is estimated using simple geometric methods as described earlier and it is superimposed on the centerline of the unworn standard profile. Then the least-squares algorithm is employed, which adjusts the position of the web portion of the worn rail shape with respect to that of the standard rail shape, so that the sum of squares of the distance between the two web portions is minimum. This is termed by the inventors a centerline superimposition technique. It should be noted herein that alternate ways of locating a worn rail shape might also be employed in various embodiments without departing from the spirit and scope of the present invention. It is also noted herein that the procedure mentioned above also establishes the coordinates of the base of the profile. In addition, the overall height of the head portion of the profile that intersects the centerline from the base center point may be calculated. It is noted that this is the only geometric feature, which may be recalculated with respect to each standard rail weight known to the system.

In a preferred embodiment of the present invention, the geometric features calculated from a worn rail shape undergoing rail-weight identification form a feature vector, which serves as input to a rail weight classifier 505. Rail weight classifier 505 uses the feature vector of the worn rail shape being identified and compares it to similar feature vectors previously acquired for every unworn standard rail shape under consideration using statistical methods. For this purpose, the geometric features for all unworn standard rail shapes are pre-calculated and stored a priori in standard rail weight repository 504, which also contains the raw data for X,Y coordinates for all unworn standard rail shapes.

Rail weight classifier 505 uses a pattern recognition and classification technique to match feature vectors of worn rail shapes to feature vectors of standard a worn rail shapes in order to determine a correct match for a particular unworn rail shape being analyzed the field. This process is achieved by using quadratic discriminant analysis.

In empirical testing, the inventors have observed that that there may be considerable variation over geometric features measured for worn rail shapes recorded over large lengths of railroad track. The amount of such variation is, in a preferred embodiment, taken into account during the process of identifying rail weights for those worn rail shapes. Geometric variation in the recorded features is estimated by calculating the standard deviation of a statistically significant sample of worn rail shapes wherein the variation for all those geometric features follow a normal distribution. More particularly, the variation satisfies an assumed, multivariate normality. In this case a quadratic discrimination rule may be employed to classify the worn rail shape.

The quadratic discrimination rule mentioned above is based on multivariate normal distribution, which takes into account the variability of all features together. The rule examines the deviation of a feature value from its standard or average value while considering its commonly occurring variation. The classification performed by classifier 505 considers, in a preferred embodiment, the sum of squares of deviations of calculated feature values from those for standard rail shapes divided by the variance (square of standard deviation, which is a measure of variation) of the feature.

Further to the above, the standard deviation values for all geometric features used are an indication of the variation of each of these features over the entire rail track. It is seen that these values do not vary considerably for the rails weights under consideration and representative default values can be arrived at to indicate the system-wide variation in typical railroad conditions. These representative values for each geometric feature used for identification are determined a priori by classifier 505 and are stored in the repository 504. Moreover, as a large number of worn rail shapes are processed, the stored values are updated to provide more realistic values. If a new rail weight is added to the repository, the default variation values are assigned to the new rail weight, until better representative values for that particular rail weight are generated over time.

In one embodiment of the present invention, a geometric feature vectors output from profile feature extractor 503 are preprocessed before classifier 505 receives them. In this embodiment, preprocessing of each vector may be performed in order to consider certain business rules governing the features that may exist. In consideration of one example, assume that a height feature of a worn rail shape is calculated to be more than that of an unworn standard rail shape. In this case according to a certain business rule, the calculated height would represent an unrealistic case and would need to be discarded by classification module 505. In such cases, a penalty function based on a sigmoid curve may be incorporated in the quadratic discrimination rule so that classifier 505 may avoid any such misclassifications. In general, a sigmoid function is used to provide a smooth variation in penalty rather than at step-like sudden variation.

In one embodiment, if a certain geometric feature of a worn rail shape is not available when comparing it with a standard rail weight stored in repository 504, the value of that feature for the corresponding standard rail shape may be assigned to it without departing from the spirit and scope of the present invention.

In a preferred embodiment, classifier 505 receives a feature vector from a worn rail shape and utilizes a quadratic discrimination rule in comparison of the received vector to vectors of unworn rail shapes to select the appropriate rail weight from the closet for a 504 that uses the least value of overall deviation from the worn rail shape.

It will be apparent one with skill in the art that although quadratic discriminant analysis is used in the preferred embodiment of the present invention, that other pattern recognition and classification techniques such as those including linear discriminant analysis or decision trees may be employed in selection of an appropriate rail weight for worn rail shape without departing for the spirit and scope of the present invention.

Once rail weight classifier 505 has determined a matching rail weight for an unworn rail profile that passes the resulting data to a post processor 506 according to a preferred embodiment. Post processor 506 validates the selection made by rail weight classifier 505 according to any existing business rules that may be applied.

It is possible that a rail weight identified using the automated identification process described above is not a correct rail weight for a particular worn rail profile being analyzed. This may occur in some cases where a worn rail shape is not recorded properly or the calculated features give an erroneous impression about the rail. Therefore, certain business rules, which may include prevalent and relevant assumptions, may be applied and used in post-processor 506 in order to validate a selection by rail weight classifier 505. One example of such a business rule may be that a single rail belonging to a rail weight A, for example, cannot be situated between to rails belonging to rail weight B. Another example might be that rails belonging to a rail weight C cannot be juxtaposed with rails belonging to a rail weight D. Still another example of the business will might be that the rail weight for the inner rail on a curve cannot be E/F/G, while that for the outer rail cannot be H/I. The described rules are merely generic examples. Post-processor 506 may consider various rules associated with railroad practices in place.

However, business rules alone cannot be relied upon for identification of rail weights. Moreover, a generalized way of commenting upon the validity of an identified rail weight is desired in the process. In addition, it is desired that profiles of worn rails that belong to rail weights that are unknown to the system be identified as ‘unknown’ rather than being classified according to the closest rail weight. Therefore, in accordance with an embodiment of the present invention, a rejection hypothesis may be included in the classification process. Such a hypothesis may, in one embodiment, be based on Chi-square distribution.

Because of a fact that the geometric features of rails along a track obey a particular statistical distribution such as the normal distribution, each feature may be described by a mean and a standard deviation along a particular length of track. Such distributions, consider normal, can be transformed into standard normal variables zero mean and a unit-standard deviation. That is to say that the standard normal variables for n number of features taken together follow a chi-square distribution with n degrees of freedom. For example, If [X₁, X₂, . . . , X_n] are n different variables with means [μ₁, μ₂, . . . μ_n] and standard deviations [σ₁, σ₂, . . . , σ_n], then a chi-square distribution with n degrees of freedom is described as follows: $Y=\sum _{i=1}^n{\left(X_i-{\mu }_i\right)}^2/{\sigma }_i^2$

Further explained, the chi-square test is a test for rejection of a hypothesis that a worn rail shape with calculated features [X₁, X₂, . . . , X_n] belongs to a rail weight with mean features [μ₁, μ₂, . . . , μ_n] with representative standard deviations [σ₁, σ₂, . . . σ_n]. The hypothesis is rejected, meaning that the worn rail shape does not belong to the rail weight under consideration if the following equation is true: $Y=\sum _{i=1}^n{\left(X_i-{\mu }_i\right)}^2/{\sigma }_i^2>{\chi }_n^2\left(0.95\right)$
where χ_n²(0.95) represents the chi-square distribution value for 95% confidence for n degrees of freedom.
ELSE it can be stated with 95% confidence that the worn rail belongs to the rail weight under consideration. The chi-square values are tabulated for different degrees of freedom and the relevant value is made available depending on the number of features extracted. The final result in output of postprocessor 506 is a final decision on rail weight classification for the worn rail shape. The final rail weight identification may be one of the rail weights from repository 504, or it may be a rail weight that is unknown to the system.

Once a correct rail weight is selected and validated, a rail wear calculator 507 is provided for calculating the amount of wear of the worn profile compared to the standard of profile of the same rail weight. Rail wear calculator 507 superimposes the worn rail shape on the unworn standard rail shape belonging to the identified rail weight. This process can be achieved by the centerline superimposition or two-dimensional cross-correlation or the pinning and refining techniques described above. In a preferred embodiment, the centerline superimposition technique is used. In this embodiment, after locating the worn rail shape exactly on top of the unworn standard rail shape, the height of the worn rail shape along the centerline is calculated. This value when subtracted from the height of the unworn standard rail shape along the centerline provides an estimate of head-wear. Likewise, the distance of the inner side of the worn rail shape at gage face (⅝ inches below the highest point) from the centerline is also calculated. This value, when subtracted from the same for the unworn standard rail shape provides an estimation of the gage face-wear. Additional measurements of wear such as area of wear, material loss, lip, required grinding profile, and so on may be calculated as required.

After rail weight identification and rail wear calculation is completed rail wear calculator 507, in a preferred embodiment, stores the rail weight identification, the wear data results, against a geographic location indicator and date and time indicators in a rail lay mapping repository 508. Geographic location is available through a global positioning system (GPS), or through manual effort. The purpose for repository 508 includes serving as a reference for application of additional tests and data on identified rail weights for rail-wear estimation performed on the same track location at future dates.

Software 500 may be provided as a software suite that runs on a computer-based platform with or without a graphical user interface without departing from the spirit and scope of the present invention. In one embodiment of the invention, software 500 may be integrated with a standalone hardware system that may be packaged provided to existing rail measurement cars for integration to existing optical components and equipment.

One with skill in the art will appreciate that software 500 may be provided with only the required components for facilitating automated rail weight identification from data input by existing optical measurement equipment and methods without departing from the spirit and scope of the present invention. Moreover, such minimum components may be integrated with an existing rail wear calculation system wherein formally manual rail weight input or selection was required. In this case, rail wear calculator 507 and, perhaps post processor 506 may be optional components.

The methods and apparatus of the present invention may be applied to and integrated with any existing rail wear calculation system that uses geometric features captured using optical equipment wherein such a system is enhanced by the invention for automated rail weight identification and further notification of identification of rail weights that may not be known to such a system. The methods and apparatus of the present invention should be afforded the broadest possible interpretation in light of the many embodiments described. The spirit and scope of the present invention is therefore limited only by the claims that follow.

Claims

1. A system for identifying rail weights applicable to worn railroad rails measured along a railroad track system comprising:

at least one track measurement device for recording at least a portion of a profile shape of the worn rails; and,

a data processor running a software implement for accepting recorded data as input data from the at least one track measurement device and for processing the input data against known unworn railroad rail data to facilitate selection and application of the appropriate rail weights equated thereto to the worn railroad rails.

2. The system of claim 1, wherein the track measurement device uses image-capturing technology to record track data.

3. The system of claim 1, wherein the track measurement device further includes audio emitting and capturing technology to record track data.

4. The system of claim 1, wherein the appropriate rail weights are identified for use in rail wear measurement of the physical wear subjected to the worn rails.

5. The system of claim 1, wherein the at least one-track measurement device is installed in a track measurement railcar.

6. The system of claim 1, wherein the profile shape of the worn rails is a transverse cross-sectional view of the rail shape.

7. The system of claim 1, wherein the software implement directs rail weight identification and rail wear calculation.

8. The system of claim 1, wherein a profile shape of a rail comprises a set of geometric features applicable and identifiable to parts of the rail profile.

9. The system of claim 8, wherein the geometric features include one of, all of, or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle.

10. The system of claim 1, wherein the unworn rail data comprises profile shapes of unworn rails, each profile including a set of geometric features making up the profile shape, the features applicable and identifiable to parts of the rail profiles.

11. The system of claim 10, wherein the geometric features include one of, all of, or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle.

12. The system of claim 4, further including a data repository for storing calculated rail-wear data results along with geographic information and date and time information.

13. A software implement for identifying rail weights applicable to worn railroad rails measured along a railroad track system comprising:

a feature extractor for extracting geometric features from recorded data of the worn rails; and

a rail weight classifier for comparing the extracted features to the same features of unworn rail shapes and for selecting an appropriate rail weight equated to an unworn rail shape for application to a worn rail.

14. The software implement of claim 13, wherein the feature extractor combines geometric feature data produce a geometric vector.

15. The software implement of claim 14, wherein the rail weight classifier compares the geometric vector against a store of geometric vectors previously created for unworn railroad rails.

16. The software implement of claim 13, wherein the geometric features derive from captured image data.

17. The software implement of claim 13, wherein the geometric features derive from captured audio data.

18. The software implement of claim 13, further comprising a preprocessor for creating X,Y coordinate pairs for feature extraction.

19. The software implement of claim 13, further comprising a post processor for validating identified rail weights applied to worn rails against a set of business rules.

20. The software implement of claim 13, further comprising a rail wear calculator for calculating the correct amount of wear subject to a worn rail compared to an unworn rail of the same identified rail weight.

21. A method for identifying a correct a rail weight for a worn rail profile given data recorded from the worn rail by a track measurement system including steps of:

(a) identifying geometric features of the worn rail profile;

(b) unifying the identified features as a single feature vector;

(c) comparing the single feature vector to one or more similarly constructed feature vectors previously stored for comparison, the previously stored feature vectors representing unworn rail profiles; and

(d) selecting a feature match of that of a worn rail profile to that of an unworn rail profile.

22. The method of claim 21, wherein in step (a), the geometric features of the worn rail profile include one of, all of, or a combination of web radius, fillet radius, profile height, head width, web width, or flange angle.

23. The method of claim 21, wherein in step (a), the geometric features are derived from image data captured from a worn rail.

24. The method of claim 21, wherein in step (a), the geometric features are expressed as a set of X,Y coordinate pairs.

25. The method of claim 21, wherein in step (a), the geometric features are discernible from one another using linear or quadratic discriminant analysis.

26. The method of claim 21, wherein in step (b), the single feature vector contains the data from all of the unified features considered in creating the vector.

27. The method of claim 21, wherein in step (d), if there is no match accomplished to a previously stored feature vector, an indication of match to a rail weight not known to the system is returned.

28. The method of claim 21, wherein in step (a), the identified geometric features are adjusted to correct for a detected angle of cant of the worn rail.

29. The method of claim 21, further including a step (e) upon obtaining a match in step (d) wherein the resulting data of the match is input into a rail wear calculator.

30. In a system for identifying rail weights applicable to worn railroad rails measured along a railroad track system, a method for correcting a worn rail profile for an angle of cant comprising steps of:

(a) detecting the existence of cant attributed to a web portion of the rail profile;

(b) retrieving a sample of any like web portion of an unworn rail aligned to true vertical;

(c) cross-correlating the web portion of the worn rail to the web portion of the unworn rail;

(d) identifying the direction of cant of the worn rail against the unworn rail;

(e) rotating the web portion of the worn rail a number of increments of angle opposite the cant direction; and

(f) accepting as a correction angle, the sum of rotation increments required to obtain a least discrepant match in cross-correlation to the symmetry of the web portion of the unworn rail profile.

31. The method of claim 30, wherein in step (a), the worn rail profile is described by a set of X,Y coordinate pairs.

32. The method of claim 30, wherein in step (a), more than one worn rail profile is considered for cant correction along a same track section.

33. The method of claim 30, wherein in step (b), the sample is retrieved from a repository previously stored unworn rail profiles.

34. The method of claim 30, wherein in step (c), cross-correlation considers vertical symmetry.

35. The method of claim 30, wherein in step (c), cross-correlation considers horizontal symmetry.

36. The method of claim 30, wherein in step (c), cross-correlation is two-dimensional.

37. The method of claim 30 further including a step (g) for estimating height of the worn rail profile.

38. The method of claim 30, wherein in step (f), a least discrepant match considers the best equation for a line of symmetry of the worn rail profile considering the minimum distance equalities of like geometric features located on either side of the line of symmetry in cross-correlation.

39. The method of claim 21 wherein, in step (d), selection is based on pattern recognition results.

40. The method of claim 39 wherein pattern recognition is based on linear or quadratic discriminant techniques and associated rules.

41. The method of claim 39 wherein pattern recognition is based on use of a decision tree.

42. The system of claim 1 wherein processing includes statistical comparison.

43. The software implement of claim 13 wherein the method for comparing is statistical comparison.

44. The method of claim 21 wherein in step (c), the method for comparing is statistical comparison.